Reader Deep Dive

Daily briefing that connects your recent reading to your long-term archive.

Security Scan

The skill's stated purpose (connect Readwise saves to your archive and send a WhatsApp briefing) is plausible, but there are multiple mismatches between what the README/SKILL.md declare and what the script actually requires and does — notably missing env declarations (TARGET_NUMBER), undocumented dependencies (gemini, clawdbot), and potential data flow to external LLMs — so you should not install without clarification and tightening of permissions.

Purpose & Capability

Requesting READWISE_TOKEN aligns with the stated Readwise integration. However the skill also depends on a WhatsApp delivery mechanism (clawdbot) and an LLM CLI (gemini) which are not declared in the registry's required binaries or env. The README lists curl, jq, clawdbot, and gemini as requirements but the registry metadata does not. This inconsistency suggests incomplete or sloppy declarations rather than a feature mismatch, but it reduces trust.

Instruction Scope

The runtime script (scripts/brief.sh) fetches full Readwise results and archive hits and pipes them into the gemini CLI (an external LLM), then sends the model-generated briefing via clawdbot to a phone number. The SKILL.md/README instruct the user to set only READWISE_TOKEN, but the script also requires TARGET_NUMBER (and implicitly expects clawdbot to be configured). The SKILL.md says it uses the "system's default LLM" but the script calls gemini explicitly — a contradictory instruction about where data is sent. The script transmits your saved items and archive contents to an external LLM process, which may leak private content to that LLM provider.

Install Mechanism

There is no install spec (instruction-only plus a shell script), so nothing is automatically downloaded or installed by the registry. That's lower risk than arbitrary downloads. However the README requires external CLIs (gemini, clawdbot) which would need to be installed by the user; those installers are not specified here.

Credentials

The registry declares only READWISE_TOKEN as a required env var, which is appropriate for Readwise access. The script however aborts if TARGET_NUMBER is unset and uses clawdbot to send messages — TARGET_NUMBER is required at runtime but is not declared. There is no declaration of any clawdbot authentication token or configuration variable; if clawdbot requires credentials those are missing from the manifest. Declaring only READWISE_TOKEN under-represents the environment/credential surface the skill needs.

Persistence & Privilege

The skill does not set always:true and is not forced into every agent run. However disableModelInvocation is not set, so the model may be allowed to call this skill autonomously. Because the skill can send WhatsApp messages and will forward Readwise contents to an external LLM (gemini), allowing autonomous invocation without tighter guardrails increases the risk of unwanted message-sending or data exposure.

What to consider before installing

Do not install or run this skill until the author clarifies and tightens its declarations. Specific steps to request or perform before using it: - Require the author to list all runtime dependencies and env vars in the registry metadata: at minimum READWISE_TOKEN, TARGET_NUMBER, and any clawdbot/gemini credentials or config. The script currently fails if TARGET_NUMBER is unset but that variable is not declared anywhere. - Confirm where the LLM call goes: the script uses the gemini CLI (sends your Readwise data to whatever provider gemini is configured to call). If you care about confidentiality, either remove/send less raw data or ensure gemini is a local/private LLM instance. The SKILL.md's "system default LLM" claim is inconsistent with the explicit gemini use. - Verify clawdbot configuration and credentials: understand which service will actually send WhatsApp messages and whether additional tokens or permissions are required; ensure you are willing for the skill to send messages to TARGET_NUMBER automatically. - Consider operational restrictions: require explicit user invocation or set disableModelInvocation:true so the model cannot autonomously trigger deliveries, and add confirmation prompts before sending messages. - Audit network/data flow: the script sends your saved items and archive hits to an LLM process and includes URLs in the payload — review this for privacy concerns. - If you still want to use it: run in an isolated environment first, with test tokens and a test phone number, and inspect gemini/clawdbot binaries to confirm they behave as expected. If the author updates the manifest to accurately list binaries and env vars, documents the data flow to external services, and offers an option to keep the LLM local and to require explicit user approval before sending messages, the inconsistencies would be resolved and the skill could be re-evaluated.

Your reading list shouldn't be a write-only memory. This skill checks what you've saved recently, finds connected ideas from your deep archive (last 3 days, 3 months, or years ago), and sends you a high-signal briefing with context on why you should revisit them.

It turns "I saved that somewhere" into "Here is the timeline of your thinking on this topic."

How it works

1. Scans Recent Saves: Checks your Readwise Reader "new" folder for the last 24 hours.
2. Identifies Themes: Uses your system's default LLM to figure out your current obsession.
3. Temporal Context: Searches your library history and finds relevant items from different timeframes.
4. Delivers Briefing: Sends a WhatsApp message with a "Deep Dive" summary connecting your current saves to your past library.

Setup

1. Get your Access Token from readwise.io/access_token.
2. Set it in your environment:

   export READWISE_TOKEN="your_token_here"
   

Usage

Manual Trigger:

bash scripts/brief.sh

Schedule (Cron): Run it every afternoon at 2 PM:

clawdbot cron add --id reader_brief --schedule "0 14 * * *" --command "bash scripts/brief.sh"

Customization

You can tweak the prompt in prompts/briefing.txt if you want a different tone or format. By default, it uses a clean, WhatsApp-friendly style.